|
Activity Number:
|
128
|
|
Type:
|
Invited
|
|
Date/Time:
|
Monday, August 12, 2002 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
International Association of Mathematicial Geology (IAMG)
|
| Abstract - #300651 |
|
Title:
|
Nonlinear Spatio-Temporal Statistics via Monte Carlo Methods Implemented in a JavaSpaces Distributed Computer
|
|
Author(s):
|
Timothy Haas*+
|
|
Affiliation(s):
|
University of Wisconsin, Milwaukee
|
|
Address:
|
P. O. Box 742, Milwaukee, Wisconsin, 53201, U.S.A.
|
|
Keywords:
|
spatial statistics ; geostatistics ; Monte Carlo method ; distributed computing ; JavaSpaces ; parallel computing
|
|
Abstract:
|
Coarse-grained parallel computing has the potential for allowing a variety of computationally intensive spatio-temporal statistical calculations to be performed by anyone with access to a network of 50 or more PCs. These calculations include robust estimation of nonlinear spatio-temporal statistical trend models, Monte Carlo assessment of model goodness-of-fit and parameter estimate reliability, optimal prediction of a spatio-temporal random field at many locations under asymmetric loss, and the construction of and high speed access to a distributed spatio-temporal statistical database. A distributed computer that performs these calculations can be constructed through a transaction space protocol using either JavaSpaces from Sun Microsystems or TSpaces from IBM. As an example, a nonlinear spatio-temporal trend model is estimated with Minimum Distance (a robust statistical parameter estimator) followed by a Monte Carlo computation of parameter estimate standard errors with a JavaSpaces program running on the 50 PCs contained in an instructional computer laboratory during hours that the laboratory is closed.
|
